イベント情報
【GIR公開セミナー】Dr. Arif Behic Tekin / エーゲ大学（トルコ）
| 日時 | 2026.2.17（13：30～15：00） |
|---|---|
| 会場 |
東京農工大学 府中キャンパス 連合農学研究科棟 4 階 第2会議室 ミーティングID：833 3400 1928 パスコード：103596 |
| 講演者 | Dr. Arif Behic Tekin |
| 所属機関 | エーゲ大学 （トルコ） |
| 講演タイトル | “AI-Integrated Framework for Dynamic Variable Rate Irrigation” ＜要旨＞ Optimization of irrigation in cotton production represents a complex, non-linear control problem characterized by high stochasticity, delayed rewards, and significant spatiotemporal variability. Traditional rule-based controllers fail to adapt to these dynamic environmental states, often leading to suboptimal water use and fiber quality degradation. This study presents an Active Cyber-Physical System (CPS) driven by a Deep Reinforcement Learning (DRL) agent designed to autonomously learn and execute texture-aware irrigation strategies. The proposed agent operates within a high-dimensional continuous state space, fusing multi-modal inputs: static soil hydraulic properties (derived from Mobile EC surveys), real-time matric potential (from micro tensiometers), and spatiotemporal plant stress metrics (UAV-based CWSI). A key innovation is the integration of a predictive look-ahead horizon, where the agent ingests weather forecast vectors to anticipate future precipitation and evaporative demand, effectively minimizing future regret in its decision policy. The DRL agent utilizes a discrete, texture-adaptive action space, learning to select between "Pulse" and "Soak" actuation modes to optimize infiltration based on local soil physics. The reward function is engineered to maximize net economic value, penalizing stress events specifically during fiber elongation and maturation phases to safeguard lint quality parameters (length and micronaire). The trained model is deployed for real-time edge inference on an NVIDIA Jetson Orin Nano, enabling low-latency, autonomous operation independent of cloud connectivity. Preliminary results demonstrate that this active learning framework outperforms conventional Model Predictive Control (MPC) in maintaining optimal canopy temperature profiles while maximizing Water Use Efficiency (WUE). Keywords: Deep Reinforcement Learning (DRL), Active Cyber-Physical Systems, Edge AI, Predictive Control, Crop Water Stress Index (CWSI), Precision Agriculture. |
| 言語 | 英語 |
| 対象 | どなたでも、ご参加いただけます。 |
| 主催 | グローバルイノベーション研究院 食料分野 加藤チーム |
| お問い合わせ窓口 | グローバルイノベーション研究院・農学研究院 加藤 亮 e-mail: taskkato ( ここに @ を入れてください）cc.tuat.ac.jp |
| 備考 | 本セミナーは、オンライン・対面型の同時開催となります。 |
このページの上部へ